Breast cancer detection

This use case cover a breast cancer binary classification using microscopic biopsy images Worldwide, breast cancer is the most-common invasive cancer in women. Along with lung cancer, breast cancer is the most commonly diagnosed cancer, with 2.09 million cases each in 2018. Breast cancer affects 1 in 7 (14%) of women worldwide. This types of algorithms can help to increase the number of diagnosis and discard the true negatives, leaving only the positives and false negatives to doctors.

Datasets

The original dataset is divided on train and test forlders with two classes, malignant and benign, but Perceptilabs has the capabilities to divide the dataset in train, test and validation, then we merge the full dataset in a single dataset with the following distribution:

imagen

You can access to the new dataset using perceptilabs github.

Model

Layer Configuration
ResNet50 include_top=false, pretrained=imagenet
Dense Activation=ReLU, Neurons=128
Dense Activation=ReLU, Neurons=64
Dense Activation=SoftMax, Neurons=2C

Workspace

Statistics view

Accuracy plot

3 Likes

I really like the way you present these example models!

1 Like

Iā€™m glad you like them @JulianSMoore :smiley:

Let us know if there is any use cases you would be interested in seeing. We are currently looking at building out some more complex models in PL and then also doing some segmentation.